headphones: understanding frequency response and THD

[RickSanchez]

Hey everyone, fairly new to ASR (big fan!) and wanted to ask about a couple of measurements for headphones:

• frequency response (FR)
• total harmonic distortion (THD)

I've been somewhat successful in finding these measurements for a number of headphones (inner|fidelity, RTings, etc.); my problem is that I don't know how to interpret these graphs. (For example, see the below measurements for FR + THD for the HiFiMan HE4xx headphones.)

I have a very basic understanding of the Harman curve vs. flat for FR, but overall I'm hoping someone could point me to a good beginner's guide for making sense of what these measurements can tell me about a headphone's sound. I'm hoping to avoid folks' subjective opinions about what measurements / headphones sound good or don't sound good. Rather, I'd like to be able to understand things like "in this range x thing is inaudible" or "these values represent a lot of bass". Not asking anyone to write a full tutorial here; if there are links to other sites that have this info that would be more than sufficient. Thanks!

 

[solderdude]

I have written some words about this subject. Way too much to post here.

https://diyaudioheaven.wordpress.com/tutorials/how-to-interpret-graphs/

When you understand all of these plots and know what measurements say what about some aspects then you can make some sort of educated guess about how it will sound when you know a few reference headphones and how they measure.
You should not compare frequency plots between different measurebators. They all differ, sometimes well over 10dB.

Let's just say that I might be able to predict how the tonal balance will be when looking at plots (and comparing them with plots of the same measurebator of known, to me, headphones).
That doesn't mean I can predict
A: how it will sound to someone else
B: How it really sounds
C: Whether I will like it or not
D: How the comfort and fit is
etc... etc...

You can view headphone measurements as an indication... but that's it.

 

[RickSanchez]

I have written some things about this subject. Way too much to post here.

https://diyaudioheaven.wordpress.com/tutorials/how-to-interpret-graphs/

Thanks @solderdude , this is great stuff.

 

[zermak]

@RickSanchez
Don't listen to @solderdude he knows nothing about headphones and electronics
You should ask to your girlfriend or wife or partner if he/she can hear a difference from the kitchen; as simple as that no matter if you using headphones, she can anyway.

 

[RickSanchez]

@RickSanchez
Don't listen to @solderdude he knows nothing about headphones and electronics
You should ask to your girlfriend or wife or partner if he/she can hear a difference from the kitchen; as simple as that no matter if you using headphones, she can anyway.

Ha!

Hmmmm, I'm currently single otherwise I'd follow your advice. I guess for now the best thing I can do is upgrade from my $75 Sony MDR-7506s to a $2500 pair of headphones. I'm sure at that price I'd be able to experience some sort of musical nirvana -- I forgot to mention I have golden ears -- that we all know is simply not possible with inexpensive pieces of audio equipment ...

 

[amadeuswus]

I have written some words about this subject. Way too much to post here.

https://diyaudioheaven.wordpress.com/tutorials/how-to-interpret-graphs/

When you understand all of these plots and know what measurements say what about some aspects then you can make some sort of educated guess about how it will sound when you know a few reference headphones and how they measure.
You should not compare frequency plots between different measurebators. They all differ, sometimes well over 10dB.

Let's just say that I might be able to predict how the tonal balance will be when looking at plots (and comparing them with plots of the same measurebator of known, to me, headphones).
That doesn't mean I can predict
A: how it will sound to someone else
B: How it really sounds
C: Whether I will like it or not
D: How the comfort and fit is
etc... etc...

You can view headphone measurements as an indication... but that's it.

An interesting article on headphone testing by Keith Howard just came out yesterday:

https://www.hifinews.com/content/better-headphone-testing

 

[RickSanchez]

You can view headphone measurements as an indication... but that's it.

And that's all I'm looking for. Something that is directionally helpful, not definitive. And hopefully useful in identifying extreme measurements that would indicate products I should avoid. (e.g., extreme bass.)

 

[daftcombo]

Even on great headphones like the Focal Clear, on the FT graph you can see huge variations (like 15dB differences). If you saw that on speakers, everybody would be screaming. Is it an artefact from measurements or is it that our ears are more picky with speakers because side-walls might amplify such variations for instance?

 

[solderdude]

The explanation below may be what you are looking for.

It gives an indication what 'sound deviations' you can expect from certain parts of the frequency range.
The higher the peak or bigger the dip the 'worse' the effect.
When there are more than 1 serious peaks or dips all the 'sound modifications' apply.
Some amplify heard affects others can 'mask' effects.

Boosted bass as well as rolled of mids + treble is the same effect but can be described by one person as bassy and by another person as muffled or lacking in clarity/brightness.
Sometimes dips aren't perceived as such when preceeded or followed by a peak (if not too big).

 

[MRC01]

I'll add that with headphones, frequency response measurements are even fuzzier because of HRTF. This is not true of loudspeakers, since they are a distant source so we each hear them through our individual HRTF. Headphones bypass the HRTF so they must simulate it, and it is significantly different for each person.

 

[daftcombo]

The explanation below may be what you are looking for.

It gives an indication what 'sound deviations' you can expect from certain parts of the frequency range.
The higher the peak or bigger the dip the 'worse' the effect.
When there are more than 1 serious peaks or dips all the 'sound modifications' apply.
Some amplify heard affects others can 'mask' effects.

Boosted bass as well as rolled of mids + treble is the same effect but can be described by one person as bassy and by another person as muffled or lacking in clarity/brightness.
Sometimes dips aren't perceived as such when preceeded or followed by a peak (if not too big).

Very useful graph. But those are 5dB peaks, -10dB dips...
On NO AUDIOPHILE the guy will use his DSP to correct even +2dB peaks or -3dB dips. Is it just a fantasy or are such small variations more audible for loudspeakers than for phones?

 

[solderdude]

It is just an indication where the dips are.
The higher the peak or bigger the dip the 'worse' the effect.
So just +2 dB is a very mild effect + 10dB is extreme. Above a certain value it becomes audible and at another level annoying.
One's 'punchy bass' is another's 'boomy bass'.
One's sibilance is another's, lovely detail.

In general, peaks are much more sound degrading than dips so one needs to be aware of peaks.
Trouble is... measurebator A measures +6dB peak where measurebator B measures a 5dB dip in the same headphone.
Only your ears can tell which measurement to believe.

 

[MRC01]

Very useful graph. But those are 5dB peaks, -10dB dips...
On NO AUDIOPHILE the guy will use his DSP to correct even +2dB peaks or -3dB dips. Is it just a fantasy or are such small variations more audible for loudspeakers than for phones?

Small amplitude variations are audible if they cover a wide enough frequency range and the music you're listening to has energy at those frequencies. I find no significant difference between headphones & speakers in sensitivity of the smallest EQ I can detect.

The HRTF is essentially a boost in the 2-5 kHz region, peaking around +15 dB (!!) This is an over-simplification but it shows what the shape of your head, outer and inner ear does to sound before you perceive it. With such large variance from person to person, in a critical frequency range. It's no wonder that equally good well trained listeners so often disagree about what headphone is most linear or natural sounding.

One might wonder how are we ever going to hear +/- 1 dB variations in EQ through all this much larger EQ and phase shift that our own heads & ears are doing to the sound. The HRTF is more or less constant, you live with it your entire life and it becomes your "normal", the lens through which you hear the world. So of course we can hear small EQ changes on top of it.

 

[solderdude]

On NO AUDIOPHILE the guy will use his DSP to correct even +2dB peaks or -3dB dips. Is it just a fantasy or are such small variations more audible for loudspeakers than for phones?

EQ'ing headphones exactly acc. to certain measurements isn't the wisest method. Exact 'compensations' such as Sonarworks and Oratory may give an audible improvement or partial improvement and partial worsening or worsens the sound as ears (and other test rigs) will (can) give VERY different results.

One should do the following: Apply oratory measurements and have it measured by Sonar works or R'tings or whomever.
Or the other way around. I am quite certain that what measures 'exactly flat' on one rig will measure horrible on another rig.
Shift the headphone a few mm and measure the 'correction again' .. peaks and dips galore.
Who is right ? Which one is 'righter' than the other ?

Should one EQ exactly to measurements or look for commonalities in different measurements and EQ these out with an average EQ ?

 

[Fluffy]

People above me wrote useful information so I won't repeat it. I'll give you my honest opinion after messing obsessively with headphone measurements and listening to a lot of headphones – the only way to know if a headphone will sound good to you, is to listen to it. No review or measurement can definitively tell you that.

Measurements of different headphones by the same measurer can help deduce that if you like one headphone there is some probability that you will like another one that measurers similarly. But this is not definite and I have been disappointed many times by headphones that measured very similarly to ones that I liked.

And also, "flat" or "good" measuring headphone does not promise you enjoyment. There are superb headphones that measures like a dream and make me wanna shoot myself, and there are headphones that measures like dog balls and I enjoy the heck out of them (Meze 99 my love). With headphones, it's so so dependent on your own ear shape and listening habits, that it gets really hard predicting things.

Given all that, I found that usually if a headphone has some part of its THD graph going above 1% in frequencies higher than 150-ish hz, it can sound harsh at high volume. We usually don't really notice distortions in the bass area, but we are more sensitive to distortions in mids and treble. It's not a perfect correlation, but so far that rule of thumb worked for me for disqualifying headphones. The ones I ended up keeping all have very low THD values above the bass area.

By the way, regarding THD I would trust Innefidlity's measurements more than any other:

https://www.innerfidelity.com/images/MassdropHiFiMANHE4XX.pdf

The HE4XX specifically gets a bit distorted at some parts of the spectrum. I personally found it to have too much treble and sometime a bit harsh sounding.

 

[MRC01]

EQ'ing headphones exactly acc. to certain measurements isn't the wisest method. Exact 'compensations' such as Sonarworks and Oratory may give an audible improvement or partial improvement and partial worsening or worsens the sound as ears (and other test rigs) will (can) give VERY different results. ...

I've been unimpressed with Sonarworks and find that EQing to get measured flat response (whether with headphones or speakers) the cure can quickly get worse than the disease. EQ can be helpful, but sometimes the only difference between medicine & poison is the dosage.

For example with the LCD-2 Fazor, the Sonarworks EQ is just too much. These headphones have a response dip centered at 4k about 1 octave on each side. Without EQ, they sound very clean and natural, but a bit warm or dead. With Sonarworks, they sound artificial - like a modern amped up over-processed recording. What I find just right is a simple parametric EQ: +3.5 dB @ 4 kHz, Q=0.67. This is just enough to offset some of the dip, make the response a bit closer to neutral, without making them sound less natural.

The point is: find something that sounds close to "right" out of the box and use EQ sparingly.

 

[RickSanchez]

... Given all that, I found that usually if a headphone has some part of its THD graph going above 1% in frequencies higher than 150-ish hz, it can sound harsh at high volume. We usually don't really notice distortions in the bass area, but we are more sensitive to distortions in mids and treble. It's not a perfect correlation, but so far that rule of thumb worked for me for disqualifying headphones. The ones I ended up keeping all have very low THD values above the bass area.

Thanks, this info is really helpful.

By the way, regarding THD I would trust Innefidlity's measurements more than any other:

Is there any additional detail you can provide about why inner|fidelity would be more trustworthy?

 

[solderdude]

Innerfidelity plotted THD.
This is far less telling than when distortion plots are used that show levels of 2nd, 3rd and 4th harmonics.
So I don't think Tyll's THD plots are more revealing.
What can be derived from Tylls THD plots is driver linearity above 90dB.

 

[Fluffy]

Is there any additional detail you can provide about why inner|fidelity would be more trustworthy?

He is just the most knowledgeable and honest one, and he's been doing this for a long time (until he retired a couple years ago, sadly). This page concentrates almost all of the measurements he's done: https://www.innerfidelity.com/headphone-measurements

And there are more that were left out of this list, probably by mistake. There is no one proof that his measurements are the best, it's just that after reading a lot of his writings I've come to trust him.

As said before, you shouldn't even take his measurements at face value as absolute truth, but as a good point of reference. His compensation curve is a bit off, so the frequency responses won't directly tell you how's the tonal balance in reality. It's best to compare between different measurements to understand the difference between each headphone. The best thing about him is that he is very consistent, and all his measurements contain the same information and the same graphs, so it's super easy to compare.

edit:

Innerfidelity plotted THD.
This is far less telling than when distortion plots are used that show levels of 2nd, 3rd and 4th harmonics.
So I don't think Tyll's THD plots are more revealing.
What can be derived from Tylls THD plots is driver linearity above 90dB.

Why would breaking it up to the different harmonics would be more revealing? I think that the overall percentage of unwanted harmonics out of the whole signal is more important, because every added harmonic is bad, doesn't matter if it's second or third etc. so it's best to know how much "dirt" the headphone adds in total.

 

[solderdude]

Why would breaking it up to the different harmonics would be more revealing? I think that the overall percentage of unwanted harmonics out of the whole signal is more important, because every added harmonic is bad, doesn't matter if it's second or third etc. so it's best to know how much "dirt" the headphone adds in total.

Lets just say I respectfully disagree. When the harmonics themselves are not important Amir might as well stop measuring FFT and only plot THD.
There are headphones with higher 3rd harmonics than 2nd harmonics. It is both unwanted but 3rd harmonics point to driver non linearities where 2nd harmonics point to different 'compression' between soundwaves from the front and rear. It thus says something about the driver which isn't seen in THD.

The good part from Tylls plots is the consistency. In his case consistently incorrect enables one to compare plots. He has a vast library, the biggest one of all I might say and indeed all are the same.
His compensation, however, is not a bit off... it is terribly off. All his plots show a 10dB peak. This surely isn't the case.
I like Tyll and value the old innerfidelity, his vast network, his fun an no BS videos and mostly agree with his findings. Great character and valuable for the headphone community. The granddad of headphone audio is a title he deserved.
Too bad he left and sold his good work/name to numbnuts that smelled more profit. Of course he knew who he sold it to and what they intended to do with Inner-fidelities reputation. I don't blame Tyll one bit as it probably allowed him to live his dream (the camper thingy)